譚 亮, 鐘守銘
(電子科技大學(xué) 數(shù)學(xué)科學(xué)學(xué)院, 四川 成都 611731)
一類具有離散時滯和分布時滯的BAM神經(jīng)網(wǎng)絡(luò)的全局耗散分析
譚 亮, 鐘守銘*
(電子科技大學(xué) 數(shù)學(xué)科學(xué)學(xué)院, 四川 成都 611731)
研究了帶有離散時滯和分布時滯的BAM神經(jīng)網(wǎng)絡(luò)的全局耗散性問題;利用Lyapunov穩(wěn)定性理論和線性矩陣不等式,通過構(gòu)造Lyapunov泛函,得到了判定具有時變混合時滯的BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)是全局耗散的新準則;通過實例仿真,表明了所得結(jié)論的有效性.
雙向聯(lián)想記憶(BAM)神經(jīng)網(wǎng)絡(luò); 時變時滯; 全局耗散; Lyapunov穩(wěn)定性理論; 線性矩陣不等式
B. Kosko[1-2]將單層單向聯(lián)想記憶網(wǎng)絡(luò)推廣到雙層雙向結(jié)構(gòu),提出了雙向聯(lián)想記憶(BAM)神經(jīng)網(wǎng)絡(luò)系統(tǒng),迄今已將BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)應(yīng)用于聯(lián)想記憶、人工智能、最優(yōu)化等方面,取得了許多成就[3-5].由于神經(jīng)元之間交換信息及信號傳輸?shù)葘嶋H過程都存在信息延遲,時滯將導(dǎo)致網(wǎng)絡(luò)系統(tǒng)的性能發(fā)生改變,從而使穩(wěn)定的系統(tǒng)變得不穩(wěn)定,甚至使系統(tǒng)的演化變得更加復(fù)雜.但是,可以通過控制時滯來優(yōu)化動力系統(tǒng)[5],所以研究有時變時滯的BAM神經(jīng)網(wǎng)絡(luò)真實地模擬現(xiàn)實處理信息,具有深遠的實際意義[3-4].
近年來,時變時滯神經(jīng)網(wǎng)絡(luò)耗散性的研究得到了越來越多的關(guān)注:文獻[6-10]研究了具有時變時滯的BAM神經(jīng)網(wǎng)絡(luò)的指數(shù)穩(wěn)定性和全局指數(shù)耗散性,文獻[11]討論了BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)平衡點的全局漸近穩(wěn)定性問題,文獻[12]通過線性矩陣不等式研究了BAM神經(jīng)網(wǎng)絡(luò)不確定時滯系統(tǒng)的魯棒耗散性,文獻[13]通過構(gòu)造合適的Lyapunov函數(shù)研究了具有無約束時滯神經(jīng)網(wǎng)絡(luò)的全局耗散性,文獻[9]利用M矩陣以及線性矩陣不等式的形式研究了常時滯的神經(jīng)網(wǎng)絡(luò)的耗散性.這些研究對神經(jīng)網(wǎng)絡(luò)全局耗散性能的分析非常重要,在工程中也有廣泛的應(yīng)用[14-18].本文通過構(gòu)造Lyapunov泛函,建立判定BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)全局耗散的新準則,以及實例仿真驗證了所得結(jié)論的有效性.
考慮如下帶有離散時滯和分布時滯BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng):
(1)
ui和vj分別表示i層和j層的神經(jīng)元狀態(tài).
(2)
其中u1、u2、σ、τ均為正實數(shù),i=1,2,…,n,j=1,2,…,m.
時滯核函數(shù)Kij,Pji:[0,∞)→[0,∞)分段連續(xù),且滿足
(3)
神經(jīng)元激勵函數(shù)hi、wj滿足如下假設(shè):
1)hi、wj在R上有界,i=1,2,…,n,j=1,2,…,m;
(4)
(5)
其中
由hi、wj滿足假設(shè)的條件可推得激勵函數(shù)gi、fj滿足如下條件:
考慮系統(tǒng)(5)的初始條件為:
其中φi(s)、φj(s)在(-∞,0]上連續(xù)且有界.
定義 1[13]BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)(5)稱為耗散系統(tǒng),如果存在緊集Ω∈Rn+m,φi(s)、φj(s)是連續(xù)有界函數(shù),使得系統(tǒng)(5)的解x(t,φ)、y(t,φ)滿足:存在T>0,對?t≥T有[xT(t,φ),yT(t,φ)]T∈Ω,則Ω是系統(tǒng)(5)的全局吸引子集.
定義 2[13]假設(shè)Ω是BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)(5)的全局吸引子集,那么系統(tǒng)(5)被稱為全局指數(shù)耗散系統(tǒng),如果存在緊集Ω1,滿足Ω?Ω1?Rn+m,使得?ξ∈Rn+mΩ1,存在M(ξ)>0,α>0,滿足
則Ω1是全局指數(shù)吸引子集,其中
引理 1[19]如果a、b是n維實向量,P是一個n×n正定矩陣,則有
2aTb≤aTP-1a+bTPb.
(6)
(7)
為了定理敘述方便,先定義一些必要的參數(shù)和相關(guān)的矩陣.令
(8)
(9)
證明 定義Lyapunov泛函
(10)
其中
對Vi(t)(i=1,2,3)沿著系統(tǒng)(5)的軌跡求導(dǎo)得
運用引理1,存在正定的矩陣P滿足
由于τ<1,σ<1,同理有
運用不等式對任意a,b∈R有2ab≤a2+b2,可得:
即得:
(14)
由引理2可得
定理 2 BAM神經(jīng)網(wǎng)絡(luò)系統(tǒng)(5)是全局指數(shù)耗散系統(tǒng),Ω1是全局吸引子集,如果Ω1滿足
證明 取泛函
對V(t)沿著系統(tǒng)(5)的軌跡求導(dǎo)可得
其中(xT(t),yT(t))T∈Rn+mΩ1.對(15)式的兩端積分可得
因此,Ω1是全局指數(shù)吸引子集,系統(tǒng)(5)是全局指數(shù)耗散系統(tǒng).
考慮如下的模型:
1) 驗證定理1,取相關(guān)的參數(shù)如下:
可以算得
令u1=0.16,u2=0.1,Λ1、Λ2為單位矩陣,經(jīng)LMI驗證:當σ值在0.1~0.9變化時,τ的最大值均可取到0.999 9;反之當τ的值在0.1~0.9變化時,σ的最大值均可取到0.999 9.所以本文的結(jié)論對時滯有較大的可行范圍.
2) 為了方便驗證定理2,依然取與定理1中相同的矩陣參數(shù),且令:
根據(jù)定理2可知系統(tǒng)(5)是全局指數(shù)耗散系統(tǒng),而且
因此Ω1是全局指數(shù)吸引子集,其中0<α<4.
[1] KOSKO B. Adaptive bi-directional associative memories[J]. Appl Opt,1987,26:4947-4960.
[2] KOSKO B. Bi-directional associative memory[J]. IEEE Trans Syst Man Cybern,1988,18:49-60.
[3] 韓力群. 人工神經(jīng)網(wǎng)絡(luò)理論、設(shè)計及應(yīng)用[M]. 北京:化學(xué)工業(yè)出版社,2002:1-139.
[4] 蔡自興,徐光佑. 人工智能及其應(yīng)用[M]. 北京:清華大學(xué)出版社,2010:1-169.
[5] 曹金玲. 時滯神經(jīng)網(wǎng)絡(luò)模型的動力學(xué)研究[D]. 南京:東南大學(xué),2006.
[6] CAO J D, DONG M F. Exponential stability of delayed bi-directional associative memory networks[J]. Appl Math Comput,2003,135:105-112.
[7] LI C D, LIAO X F, ZHANG R. Delay-dependent exponential stability analysis of bidirectional associative memory neural networks with time delays:an LMI approach[J]. Chaos, Solitons and Fractals,2005,24:1119-1134.
[8] LIANG J L, CAO J D. Exponential stability of continuous-time and discrete-time bidirectional associative memory neural networks[J]. Chaos, Solitons and Fractals,2004,22:773-785.
[9] LIAO X X, WANG J. Global dissipativity of continuous-time recurrent neural networks with time delay[J]. Phys Rev,2003,E68(1):1-7.
[10] LOU X Y, CUI B T. Absolute exponential stability analysis of delayed bi-directional associative memory neural networks[J]. Chaos, Solitons and Fractals,2007,31:695-701.
[11] ARIK S, TAVSANOGLU V. Global asymptotic stability analysis of bidirectional associative memory neural networks with constant time delays[J]. Neurocomputing,2005,68:161-176.
[12] SERON M M, HIL D J, FRADKOV A L. Nonlinear adaptive control of feedback passive systems[J]. Automatica,1995,31(7):1053-1060.
[13] SONG Q K, ZHAO Z J. Global dissipativity of neural networks with both variable and unbounded delays[J]. Chaos, Solitons and Fractals,2005,25:393-401.
[14] LIAO X X, LUO Q, ZENG Z G. Positive invariant and global exponential attractive sets of neural networks with time-varying delays[J]. Neurocomputing,2008,71:513-518.
[15] ZHANG Q, WEI X P, XU J. Global exponential convergence analysis of delayed neural networks with time-varying delays[J]. Phys Lett,2003,A318:37-44.
[16] ZHANG J Y. Globally exponential stability of neural networks with variable delays[J]. IEEE Trans Circ Syst I,2003,50(2):288-291.
[17] TU Z W, JIAN J G, WANG B X. Positive invariant and global exponential attractive sets of a class of neural networks with unbounded time-delays[J]. Commun Nonlinear Sci Numer Simul,2011,16:738-745.
[18] CAO J D, CHEN T P. Globally exponentially robust stability and periodicity of delayed neural networks[J]. Chaos, Solitons and Fractals,2004,22(4):957-963.
[19] XU S, LAM J, HO D W C, et al. Global robust exponential stability analysis for interval recurrent neural networks[J]. Phys Lett,2004,A325(2):124-133.
[20] BOYD B S, GHAOUI L E, FERON E, et al. Linear Matrix Inequalities in Systems and Control Theory[M]. Philadephia:SIAM,1994.
2010 MSC:34D20; 92B20
(編輯 周 俊)
Global Dissipativity of a Class of BAM Neural Networks with Discrete and Distribute Time Delays
TAN Liang, ZHONG Shouming
(SchoolofMathematicsScience,UniversityofElectronicScienceandTechnologyofChina,Chengdu611731,Sichuan)
In this paper, we investigate global dissipativity for BAM neural networks with discrete and distribute delays. The Lyapunov stability and the linear matrix inequality (LMI) approach are employed in our work. By establishing a new Lyapunov functional, we obtain novel criteria for global dissipativity analysis. Some numerical examples are given to illustrate the effectiveness of the proposed method.
bi-directional associative memory; time-varying delays; global dissipativity; Lyapunov stability method; linear matrix inequality
2015-10-26
國家自然科學(xué)基金(61273015)
O175.13
A
1001-8395(2017)01-0011-07
10.3969/j.issn.1001-8395.2017.01.002
*通信作者簡介:鐘守銘(1955—),男,教授,主要從事動力系統(tǒng)與控制論的研究,E-mail:zhongsm@uestc.edu.cn